Study On Power System Transient Stability Assessment Based On Improved Long Short-term Memory Neural Networks

Power system transient stability assessment is the main link of power system dynamic security assessment and the key technology of early warning.With the increasing degree of grid interconnection and the continuous growth of new energy grid connection,the requirement of the accuracy and precision of transient stability assessment are further improved,and traditional methods are facing many challenges in this regard.The development and application of deep learning technology in power system provides a new way for transient stability assessment to seek a breakthrough in the accuracy and precision of assessment.After the simulation of the power system transient process,the characteristics of the variation of the power flow and the generator speed deviation in the transient process are analyzed.At the same time,combined with the existing feature selection strategy,a feature selection method based on the time series of power flow and generator speed difference as the initial feature is proposed.Combined with the time sequence characteristics of the initial characteristics,a power system transient stability evaluation model based on LSTM-DNN model is proposed.The front-end of the model uses long short term memory(LSTM)to extract the implicit features in the initial features,while the back-end uses deep neural networks(DNN)to classify and identify the transient stability results according to the features extracted from the front-end.In this model,long-term and short-term memory neural networks and deep neural networks are fused by linear superposition.In the process of training,random dropout technology,L1 and L2 regularization technology are introduced.The advantage of the model lies in its strong ability of feature extraction for time series data,outstanding ability of fitting nonlinear functions,and good generalization ability.In addition,In order to reduce the influence of artificial setting of hyper-parameters on the accuracy of model evaluation,bayesian optimization algorithm is introduced to realize the automatic optimization of hyper-parameters.Finally,the New England 39-bus system is used to verify the model.The accuracy of the new model is significantly improved compared with the traditional machine learning model,and it is better than the general deep learning model.